LISA single subject analysis using prewhitening (vlisa_prewhitening)

The program vlisa_prewhitening performs a single-subject analysis including a correction for multiple comparisons using the LISA algorithm [2018_Lohmann]. The program expects one or more preprocessed 4D fMRI files as input. The preprocessing pipeline must include a correction for baseline drifts. The files typically represent runs of an experiment acquired within the same session.

The output is a map thresholded such that FDR < alpha for every voxel. The default is alpha=0.05. The resulting image shows (1-FDR) so that larger values indicate higher significance.

Temporal autocorrelations are accounted for using the prewhitening approach [2002_Worsley].

Design files in txt-format are required as input. The number of design files must match the number of runs. The design files should not include any hemodynamic modelling. The design file is a txt-file containing one row per trial. Each row must have four values. The format of each row is: label, onset in seconds, duration in seconds, amplitude. The labels are integers starting from 1, and represent the label (event type) of that trial. For example, if there are three different event types, then their labels should be one of 1,2,3. For a detailed description see designformat_first.

The user must supply a contrast vector whose dimension matches that of the design files.

Hemodynamic modelling is specified using the parameter ‘-hemo’. Possible values are “gamma_0” (canonical function), “gamma_1” (canonical function plus its first derivative), “gamma_2” (canonical function plus its first and second derivatives), or “gauss” (Gaussian function).

The option ‘-col1’ determines whether the hemodynamic modelling procedure should add a a first (leftmost) column of the design file containing the constant value ‘1’. This column represents a global offset. If set to ‘true’ (default) then the contrast vector must contain an additional first entry which is usually set to ‘0’.

A txt-file containing additional nuisance covariates (e.g. motion parameters) may be added as a separate file. It must have one line per time volume and one column per covariate. The covariates are not subjected to any form of hemodynamic modelling, and are exempt from random permutations. These nuisance covariates need not be specified in the contrast vector. Nuisance files from several runs must be concatenated into one single file. The number of lines in this file must match the total number of volumes of all runs. By default, the mean of each regressor is subtracted (‘-demean true’).

Optionally, the design file can be obtained as a txt-file using the option ‘-plot’. The txt-file that will be output using this option contains one column for each regressor. The task regressors are convolved with a hemodynamic model, the nuisance covariates are normalized. The file can be visualized using various plotting tools (e.g. vdesignplot or gnuplot).

Note that a region-of-interest mask is required. The mask should exclude non-brain voxels, and it may cover the entire brain. In the example below, the mask is in the file “braimmask.nii”.

Caution: The preprocessing pipeline must include a correction for baseline drifts.

Examples:

vlisa_prewhitening -in run_*.v -design des_*.txt -contrast 0 1 -1  -mask brainmask.nii -out result.v -plotdesign X.txt
vdesignplot -in X.txt

Sometimes, it may be useful to obtain an output image that is not FDR-thresholded at a pre-defined alpha level. This can be achieved by setting the alpha-parameter to 1, i.e. ‘-alpha 1.0’. The resulting image can later be thresholded using the program “vlisa_applythreshold”. Example:

vlisa_prewhitening -in run_*.v -design des_*.txt -contrast 0 1 -1  -mask brainmask.nii -out result.v -alpha 1.0
vlisa_applythreshold -in result.v -out thresholded.v -threshold 0.05

Note that ‘vlisa_prewhitening’ also accepts input images in Nifti format (“run_*.nii” or “run_*.nii.gz”), but the output is always in vista format. To convert the output to the Nifti format, use the following command:

vnifti -in result.v -out result.nii

Parameters of ‘vlisa_prewhitening’:

-help

Prints usage information.

-in

Input file(s).

-out

Output file.

-design

Design files.

-contrast

Contrast vector.

-nuisance

Nuisance regressors (optional).

-demean

Whether to subtract mean in nuisance regressors. Default: true

-plotdesign

Filename for plotting design matrix X (optional).

-order

Order of AR model. Default: 1

-hemo

Hemodynamic model [ gamma_0 | gamma_1 | gamma_2 | gauss ]. Default: gamma_0

-col1

Whether to add a constant first column. Default: true

-alpha

FDR significance level. Default: 0.05

-perm

Number of permutations. Default: 5000

-mask

Region of interest mask.

-minval

Signal threshold.

-seed

Seed for random number generation. Default: 99402622

-radius

Bilateral parameter (radius in voxels). Default: 2

-rvar

Bilateral parameter (radiometric). Default: 2.0

-svar

Bilateral parameter (spatial). Default: 2

-filteriterations

Bilateral parameter (number of iterations). Default: 2

-cleanup

Whether to delete isolated voxels. Default: true

-j

Number of processors to use, ‘0’ to use all. Default: 0

References

[2002_Worsley]

Worsley et al. (2002). A general statistical analysis for fMRI data. NeuroImage 15(1):1-15. (link)

[2018_Lohmann]

Lohmann G., Stelzer J., Lacosse E., Kumar V.J., Mueller K., Kuehn E., Grodd W., Scheffler K. (2018). LISA improves statistical analysis for fMRI. Nature Communications 9:4014. (link)